Modelling palaeoecological datasets using a state-space approach: two case-studies

Quinn Asena, Jack Williams, and Tony Ives

2025-08-08

People


  • Jack Williams
  • Tony Ives
  • Angie Perotti
  • Nora Schlenker
  • David Nelson
  • Bryan Schuman
  • Jonathan Johnson
  • Vania Stefanova

multinomialTS background


A state-space approach to modelling multinomially distributed community data.


  • ESA 2023: presented early development
  • ESA 2024: ran workshop on how to fit multinomialTS
  • ESA 2025!: presenting developed use-cases
  • Do you have multinomially distributed data?
  • I’ll try to leave a couple of extra minutes for discussion.

Beyond pattern recognition


The cutting edge in palaeoecology is to establish potential relationships between patterns observed in species relative abundances and environmental covariates. For example, are observed patterns driven by:

  • Species interactions?
  • Climate variability?
  • Fire regime?

This is what we want to know if we are to use palaeoecology to inform management of contemporary ecosystems or inform potential future ecosystem states. No easy task!

State-space modelling


State-space modelling goes beyond descriptive approaches and attempts to estimate:


  • Autoregressive / density dependent processes
  • Interspecific interactions
  • Species-environment interactions
  • Combinations of the above

multinomialTS

  • Models a multinomial distribution (i.e., count data) directly
  • Accepts multiple covariates of different type
  • Incorporates process error and observation error
  • Asena et al., in review

Case studies


Tulane

  • Florida
  • ~60,000 year-long record
  • Centennial to millennial dynamics
  • Covariates
    • fungal spores (proxy for megaherbivory)
    • \(CO_2\) \(\delta18O\) (climate)
    • Heinrich events (climate-related)
    • charcoal (fire)
  • Williams et al., in prep, Grimm et al. (1993); Grimm et al. (2006)

Sunfish

  • Pennsylvania
  • ~13,000 years
  • Decadal to centennial dynamics
  • Covariates
    • lake level (proxy for humidity)
  • Johnson et al., in review

Fitting process


  • Set up multiple working hypotheses
  • Choose focal taxa
  • Choose window-span/prediction resolution
  • fit model to hypotheses or components of each hypothesis

Tulane variables

Fitting Tulane


Questions/hypotheses:

  • are biotic interactions or climatic variability the primary drivers of change?
  • is the holocene period significantly different to the full 60,000 years?

Parameter selection

  • window span / prediction resolution 200 years
  • species/taxonomic group selection
    • two functional groups
    • two key dominant species
  • interactions estimated:
    • no interactions
    • pine - oak

Tulane results

Supported hypotheses: given the data, species interactions are important but climate covariates have a stronger influence.

Sunfish description

Fitting Sunfish


Questions/hypotheses:

  • Are local conditions (humidity) a stronger driver of change than species interactions?

Parameter selection

  • window span / prediction resolution 100 years
  • species/taxonomic group selection
    • five most dominant species
    • Eastern white pine, hemlock, beech, oak, and birch
  • interactions:
    • estimated between pine and hemlock (for this example)

Sunfish results

Supported Hypotheses

Comparison of results

Results are not causal but test multiple hypotheses

Thank you for listening!

References

Grimm, Eric C., George L. Jacobson, William A. Watts, Barbara C. S. Hansen, and Kirk A. Maasch. 1993. “A 50,000-Year Record of Climate Oscillations from Florida and Its Temporal Correlation with the Heinrich Events.” Science 261 (5118): 198–200. https://doi.org/10.1126/science.261.5118.198.
Grimm, Eric C., William A. Watts, George L. Jacobson, Barbara C. S. Hansen, Heather R. Almquist, and Ann C. Dieffenbacher-Krall. 2006. “Evidence for Warm Wet Heinrich Events in Florida.” Quaternary Science Reviews 25 (17): 2197–2211. https://doi.org/10.1016/j.quascirev.2006.04.008.